Pyyan / News / 5 November 2026

SafetyMETR · Anthropic

METR estimated Anthropic's benchmark scores have grown 1.9 times faster since October 2025

1.9xgrowth rate of Anthropic's benchmark scores since October 2025, 95% interval 1.5 to 2.8x

METR published an assessment of AI research and development at Anthropic, based on access from 12 August to 29 October. Benchmark scores for Anthropic's models have grown at 1.9 times the earlier rate since October 2025, which is roughly two years of past progress in one. METR says training compute growth did not cause it and guesses most of it comes from using AI in R&D. Anthropic redacted most of the evidence METR found most informative, and METR says so.

METRAPtwo sources

Related

← All the news, newest first